Robust Video Watermarking Using Secret Sharing, SVD, … · Robust Video Watermarking Using Secret Sharing, SVD, DWT and Chaotic ... Despite many watermarking methods exist on combination

Received: August 10, 2017 171

International Journal of Intelligent Engineering and Systems, Vol.11, No.1, 2018 DOI: 10.22266/ijies2018.0228.18

Robust Video Watermarking Using Secret Sharing, SVD, DWT and Chaotic

Firefly Algorithm

Bhargavi Latha S1* Venkata Reddy Dasari2 Damodaram Avula3

1Department of Computer Science and Engineering,

Gokaraju Rangaraju Institute of Engineering and Technology, Hyderabad, India 2Department of Electronics and Communication Engineering,

Mahatma Gandhi Institute of Technology, Hyderabad, India 3Department of Computer Science and Engineering, Sri Venkateswara University, Tirupathi, India

* Corresponding author’s Email: [email protected]

Abstract: Watermarking is the process to protect and discourage the illegal sharing of multimedia content through

digital data sharing websites over INTERNET and also to prove ownership of the multimedia content. Though

various watermarking solutions are available to safeguard the digital media, at the same time many methods exist to

weaken the strength of watermark hence ownership is disproved and can be shared illegally. Strength of watermark

can be increased to make it to robust against these methods at the cost of perceptual quality. So, a compromised

approach is required. This paper proposes a video watermarking solution to maintain robustness as well perceptual

quality through optimization. This method adds secretly shared watermark bits to singular values of the discrete

wavelet coefficients with proper scaling factor, which is selected by using the Chaotic Firefly Algorithm

optimization method. Despite many watermarking methods exist on combination of Discrete Wavelet Transform

(DWT) and Singular Value Decomposition (SVD), these are not effective against all filtering attacks as these are not

using all frequency coefficients either for embedding or secret sharing. This method generates secretly shared

watermark based on singular values in DWT domain to make the system more robust against filtering attacks and

video compression techniques. This solution leverages chaotic firefly algorithm to compromise robustness and

perceptual quality of the input along with SVD and DWT. The performance of method is measured by evaluating it

on a dataset comprised of 140 videos of various genre and compared its performance with state of art methods.

Eventually, proved the method is performing well when compared to state of art methods with respect to robustness

as well as quality at the cost of computation. The stated results in experimental section shows us that watermark can

be retrieved even if watermarked audio undergoes several attacks like compression at lower bit rates, frame drop

attack and resize attack etc.

Keywords: Singular value, Wavelet-transform, Watermark, Video-watermark, Attacks, Chaotic firefly, Optimisation.

1. Introduction

In day to day life, INTERNET becomes the most

prominent channel to share the multimedia content.

This opportunity brings the both pros and cons to

the digital media industry. Positive side of this

increases the revenue of the company. In contrast,

unfair distribution can be termed as infringement of

the multimedia data over INTERNET brings huge

loss to the company, multimedia data could be an

audio, an image or a videos etc. These cons attracted

lot of researchers to find different ways to mitigate

this and came up with a solution called

watermarking a data before actual distribution.

In watermarking, ownership information is

inserted into the multimedia content as a watermark

and then retrieved it when necessary to prove its

ownership. But watermark can be removed by using

signal processing operations generally called as

attack, when watermark is inserted using simple

watermarking methods. Therefore, a watermarking

method should be robust enough against attacks like



filtering, scaling, cropping, rotation, or frame

dropping, etc. while maintaining the acceptable

signal quality. In this paper we propose a robust

method to watermark a video and maintain its signal

quality with the help of optimization methods.

Many video watermarking techniques exist with

various design trade-offs with the aim that

watermark should sustain even when watermarked

data is manipulated using various signal processing

operations such as filtering, scaling, cropping etc..,

without changing signal characteristics of the host

video signal. Having said that several methods exist

in the current market , which uses Discrete wavelet

transform(DWT) [1], singular value

decomposition(SVD) [2] because of its advantages

like compacts the maximum energy of the signal in

a few singular value coefficients by optimal matrix

decomposition [3] and also small distortion over the

image will not perturb the singular values

significantly. This property brings application of the

SVD into image watermarking applications and

for the same we have also adopted SVD to video

watermarking method proposed in our approach and

adopted choatic firefly algorithm to find a scaling

factor which maintains robustness and signal

quality.

2. Related work

Although, multiple works on DWT for video

watermarking exist, each method had its own

advantages and disadvantages. Few methods which

used DWT for watermarking are Hongmei Liu [4]

embedded the watermark directly to the DWT

coefficients of LL band of three level DWT, in

addition this method used the BCH error correcting

codes and also applied 3-D interleaving in order to

reduce burst errors. This method may fail against

high pass filtering since watermark was inserted in

low frequency LL band. Modifying least significant

bits of DWT coefficients makes system less

robustness against filtering attacks such as

sharpening and Gaussian filtering. Rathore[5]

followed the same approach by changing scrambling

method. H. Tian [6] applied 1D wavelet transform

on two consecutive frames, then low frequency part

was partitioned into equal size blocks to embed

watermark bit based on average pixel value of the

block and then same procedure was employed to

detect the watermark. Based on published results

this method worked well for various compression

ratios of mpeg compression, Gaussian noise but will

not work against frame drop attack as watermark bit

was embedded using two consecutive frames in

DWT domain. CE Want [7] embedded two zero

mean normally distributed watermarks in Temporal

Wavelet Transform (TWT) domain to avoid block

effects. This method inserted watermark bit into one

block of 32 frames in TWT domain. Watermark

extraction becomes challenging against frame drop

attacks. Few more approaches used DWT along with

other transforms, Method in [8] used both DWT and

SVD on both input video frame and input watermark.

This method requires to store U and V components

of watermark to retrieve the watermark during

watermark extraction and it is possible to retrieve

watermark by keeping other's watermark S

component with existing U and V components, this

in turn causes false positives.

Allali [9] employed a watermarking scheme

based on Walsh Hadamard Transform (WHT) and

DWT, where each video frame was segmented into

cubes, DWT is applied followed by 3-level WHT on

each row to embed watermark bit to high frequency

coefficients. Watermark was retrieved by using the

prediction method as specified in [10] and used

Wiener filtering. But not evaluated against frame

drop attack as well as video compression techniques

like x264 etc. Some more methods adopted

combination of more than two signal processing

techniques like DWT, DCT and SVD in [10] for

watermarking the ultrasound signal, watermark was

inserted into SVD coefficients of ultrasound in

DWT and DCT domain. This paper also employed

SVD to watermark due to its characteristics. A

concept called Binary Particle Swarm Optimization

(BPSO) used [11] to know which frames are suitable

for watermarking based on fitness value computed

from each frame, the drawback of this method is, it

should store list of frame numbers to retrieve the

watermark. Other transform like contourlet

transform was used in [12] along with discrete

wavelet transform and singular value decomposition

to watermark a given video and log polar transform

was used during extraction of watermark in order to

resist geometric attack. Li [13] proposed a method

which used discrete cosine transform in order to

select low frequency coefficients to embed a

watermark and also logistic chaotic map and error-

correction coding were adopted to make system

robust against attacks. A method [14] developed a

watermarking system which used singular value

decomposition to insert watermark and also

exploited the mosaic from all video frames to insert

a double signature in order to improve embedding

capacity. Sundararajan [15] inserted watermark in a

video by partitioning it into number of frames, then

watermark image was sliced into bit planes and

permuted them in order to embed into the segmented

shot, but this method fail against crop attack.



Method in [16] exploited discrete wavelet

transform to generate a key with the help of

watermark and binarized low frequency part of the

video frame and same is applied on every frame to

extract a corresponding key. These keys were used

during extraction process. This method fails against

filtering attacks since these attacks result in wrong

key. Block classification and visual cryptography

was used by [17] to embed watermark, where a

watermark signal was split into small watermarks

based on number of video frames in each shot. Each

small watermark was used to generate owner's share

which was inserted. This method fails to extract

watermark against frame drop attack and also this

requires synchronization method to find start

position of small watermark.

Therefore, we propose a method to eliminate

these disadvantages using multilevel DWT and

singular values obtained using SVD. Complete

watermark is inserted into single frame rather than

distributing into multiple frame to overcome the

frame drop attack, hence this method eliminates the

usage of synchronization codes to locate starting of

the watermark bit. The main contribution of this

paper when compared to state of art is computing of

four singular value matrices of four sub-bands of

DWT instead of one sub band in three level DWT to

make it robust against filtering attacks and the way

we applied secret sharing to singular values, so that

watermark can be retrieved successfully even

though two sub bands are filtered out. In secret

sharing[18] watermark will be expanded based on

host image which makes watermark very secure and

also each watermark bit is represented by four bits

so that watermark bit can be recovered efficiently

even few bits are corrupted at the cost of watermark

embedding capacity. We also used chaotic firefly

method to select best scaling factor to achieve

robustness as well as to preserve the quality of the

watermarked video. We discuss the detailed

embedding and extraction process in forthcoming

sections.

3. Background

This section discusses briefly about the discrete

wavelet transform, singular value decomposition

and chaotic firefly.

3.1 Discrete wavelet transform

Discrete wavelet transform is being used widely

in 1D and 2D signal processing due to its

advantages in signal and image processing

applications like compression, de-noising, texture

analysis and etc. Later on, its usage is carried into

watermarking. Discrete wavelet transform (DWT)

unlike DFT or DCT [19] represents a given signal

with set of basic functions efficiently and flexibly by

using filter banks, these basis functions are termed

as wavelets. In image processing domain, discrete

wavelet transform represents a given image with

series of wavelets in multi-resolution manner for

effective analysis, also signal can be viewed or

analysed both in spatial domain and frequency

domain simultaneously. A 2D DWT can be

implemented by applying 1D wavelet along the

rows and then along the columns to result in 4 sub-

bands, each contains specific range of frequency

coefficients. These can be used for any kind of

applications and as well for watermarking, which

uses these sub-bands either for inserting the

watermark bits directly or for further processing

[20]. Fig. 1 shows one level DWT filter design used

for decomposition and Fig. 2 shows the example

output of 3-level DWT.

In Fig. 1, LPF represents the low pass filter

whereas HPF represents conjugate filter such as

Figure.1 Represents DWT decomposition

Figure.2 Sample 3-level DWT decomposition



high pass filter and together can be termed as filter

banks. In this work, Haar Wavelets are selected due

to its simplicity in-terms of operations and it

decomposes the signal into four sub band signals.

Hungarian mathematician Alfred Haar introduced

the Haar wavelets, which is similar to step function.

Original signal can be computed by inverse

operation of the decomposition filter and is a

symmetry of the decomposition filter shown in Fig.1.

Instead of down sampling where up-sampling will

be used on the given four sub band inputs. All

though DWT is old transform domain, still it has

been being used in image or video watermarking

due to its advantages.

3.2 Singular value decomposition

Singular value decomposition (SVD) is

generally used for decomposing the image into sub

matrices for removing redundant data in

compression applications and also used for

watermarking. As name suggests the Decomposition

results in three matrices and they are left, right

singular vector matrix and diagonal matrix. The

diagonal Matrix consists of singular values along its

diagonal in decreasing order, where singular value

represents the energy of the given signal. These

singular values plays an important role in

compression and as well as watermarking. One

peculiar property of the singular values is small

perturbation over signal, these values are not

effected much and vice versa. Hence, these are used

for watermarking [21].

3.3 Chaotic firefly

Chaotic firefly algorithm [22] is used to find the

proper scaling factor to scale the watermark bit

during embedding phase such that robustness and

perceptual quality will be maintained by doing

optimization over given parameters. Chaotic firefly

algorithm is an extension of firefly algorithm (FA)

in order to improve the efficacy. This enhancement

can be done by replacing the random parameters and

constants of the FA. Hence, optimal value of the FA

can be obtained by using a logistic regression.

Logistic mapping is used to replace the parameters

and to improve algorithm performance as given Eq.

(1)

𝑋𝑖 = 𝑋𝑖 + 𝐼0 𝑒𝛾𝑟2

(𝑋𝑗 − 𝑋𝑖 ) + 𝛼 𝑈𝑖 (1)

Firefly algorithm is population based algorithm,

in which each member of the population is a

candidate solution of the problem that is going to be

solved. In Eq. (1), Xi represents the candidate

solution; Io represents light intensity at the source

and also can be termed as attractiveness at the

source that depends on distance r, fixed light

absorption coefficient γ. And α represents a step

size scaling factor with respect to Ui randomization

parameter. Where in above Eq. (1) first term

conveys about position of the ith firefly, second term

conveys to social component of moving the firefly i

towards the more attractive firefly j with a

component α light absorption coefficient of the

medium, third term is to represent randomised move

of the ith firefly with in the search space. During the

evaluation of chaotic firefly algorithm, it tries to

calculate value of the objective function for each

candidate solution.

This work adopted scrambling method [23] to

scramble the watermark for eliminating burst errors,

which are resulted from severe attack over

watermarked video frame using various signal

processing methods. Watermark is inserted into

frame after scrambling, when burst errors are

occurred during watermark extraction process, these

burst errors can be converted to single bit errors

using de-scrambling. Hence, these can be corrected

easily by using any error correcting codes.

In order to make the watermark more secure,

watermark can be shared among multiple samples or

each bit of watermark can be represented with

multiple bits. This can be achieved by using secret

sharing method [24]. The image to be secreted will

be represented with n-shares using a pre-determined

code book. Few shares are enough to retrieve the

original image. This paper have employed a method

[24] for generating the share image from the

watermark, each bit of watermark and feature vector

are extracted from the cover image which has been

used to generate share bits of corresponding

watermark bit. So, if watermark size is 30x30 then

share image size will be four times the watermark

size because each bit is represented by four bits. In

general, this resultant share is called private share

(p-share image) and will be inserted into video as a

watermark. During the watermark extraction, same

process is employed on watermarked video to

extract the public share image called as c-share. This

c-share and extracted p-share from watermark

extraction process will be used to extract watermark.

Dilation or erosion morphological operations can be

used to extract exact watermark.



4. Methodology

This section deals with watermark embedding

and extraction processes along with corresponding

block diagrams.

4.1 Watermark insertion

Methodology involved in embedding a

watermark into a video is given in Fig. 3. The input

video is fragmented into frames. The blue (B) and

red (R) colour channels of each frame are selected

for watermarking due to their visibility

characteristics. Same watermark approach is used to

watermark each R and B channels (components)

separately for improving the retrieval accuracy at

the cost of computation. Therefore watermarking

procedure with one component is explained here.

To generate share image, colour component is

fragmented into blocks of size 32x32 and then 3

level Discrete Wavelet Transform is applied on each

block, This results in four subbands “LL, LH, HL

and HH” per each block, where LL component

(approximate component) at the output of one level

Discrete Wavelet Transform will be used as input to

the next level Discrete Wavelet Transform except

for first level DWT where input block of an image

will be used as input. So, 3-level Discrete Wavelet

Transform results in a 4x4 size of four sub bands,

let’s say LL, LH, HL, HH sub bands. Singular Value

Decomposition is applied on four sub bands

individually and resulting a three sub matrices for

each sub band. Out of three, one is diagonal matrix

with singular values along the diagonal direction in

decreasing order and other two are said to be right

singular vector matrices and left singular vector

matrices. The singular value matrices of four sub-

bands and each bit of watermark are used to

generate share image called p-share based on the

table in [24]. Following this, row column

transformation is applied on share image for

scrambling the watermark in order to overcome the

problem of consecutive errors, this helps in

correcting single bit errors more easily than bunch

of consecutive errors by using error correcting codes.

Resultant share image size is double (2kx2k) than

input watermark size(kxk) because each input

watermark is shared to four bits so that if at all one

or two bits are corrupted due to attacks, still the

original bit can be recovered based on majority of

the four bits during watermark extraction process.

This share image is in the form of binary image and

this is converted into bipolar form i.e., matrix

having positive ones and negative ones during

inserting process.

Each bit of p-share will be inserted into one

block of the colour component in Discrete Wavelet

Transform and Singular Value Decomposition

domain during embedding process with proper

scaling factor. Each colour component is

fragmented into sub blocks of size 8x8 and then one

level Discrete Wavelet Transform is applied,

resulting four sub bands of LL, LH, HL and HH of

size 4x4. In Discrete Wavelet Transform, we used

Haar filter bank due to its simplicity in computation

and efficiency. Out of four sub-bands, Singular

Value Decomposition is performed on LL

(approximate coefficients) band thus resulting in a

three sub matrices such as left singular vector matrix,

diagonal matrix has singular values along its

diagonal direction in decreasing order, right singular

vector matrix. Out of three matrices, singular matrix

is selected for watermark inserting purpose because

minor change in singular values does not affect the

quality of the video as well as even for attacks on

video may not affect the singular values. Each bit of

p-share is added to first singular value of the

diagonal matrix based on the Eq.(2) with preselected

scaling factor, which is used to compromise the

robustness as well as perceived quality of the video.

This process is continued on each block till all the

bits of p-share are embedded. The Input video

resolution is selected such that all bits of p-share are

accommodated into a single frame. This opportunity

brings the strength to our method in case of frame

drop attack when compared to state-of-art method.

Frame drop attack is an attack performed by

dropping few video frames in between the video to

make the watermark extraction system to fail in

extracting the watermark. The dropped frames may

be consecutive or random frames throughout the

video. Where p-share is converted into bipolar form

before embedding.

𝑆1 = 𝑆1 + 𝛼(𝑠ℎ𝑎𝑟𝑒𝑏𝑖𝑡) (2)

Where S1 is first singular value of the singular value

matrix obtained from SVD decomposition and α is

the scaling factor, which will be obtained using

chaotic firefly algorithm and sharebit is the

watermark bit obtained from watermark and input

video frame. Once the watermark bits are embedded

into a colour components, reverse operation of

Singular value decomposition that is multiplication

of the modified singular value matrix(S) with left

singular vector matrix(U) and right singular vector

matrix(V) in order (U*S*V') and then inverse

Discrete Wavelet Transform will be performed

along with modified LL sub band and remaining sub



Figure.3 Watermark inserting process into a video

Figure.4 Watermark extraction process from a video

bands such as LH , HL and HH to get back the

original watermarked component and then these

three components R, B, and G components are

combined to form a watermarked frame. The same

procedure is applied on B colour component of the

video frame and four subbands of each block to

make it robust against filtering attacks. Finally, all

these frames are combined on the fly using

FFMPEG [25] to get back the watermarked video.

4.2 Watermark extraction

The block diagram for watermark extraction is

shown in Fig. 4. Watermark is extracted by using

both original video and watermarked video. Where

R and B components are selected because these

components are used for embedding the watermark.

From watermarked R component, c-share image is

generated by using same procedure applied during

p-share generation and the same from B component

as well. Embedded watermark is extracted by

applying one level Discrete Wavelet Transform, and

Singular Value Decomposition on LL component on

both R components of input video as well as

watermarked video and first singular values of both

frames are compared based on majority either one or

zero is extracted. The extracted watermark is same

size as c-share image. Finally extracted watermark

and c-share are used to extract original watermark

by performing exclusive-or operation and by using

majority voting from block of four bits. Still if any

error persists then those can be eliminated by using

error correcting codes and by using morphological

operations.

4.3 Chaotic firefly algorithm

Chaotic firefly optimisation method should be

populated first like other optimisation methods. For

this, 15 attacks on watermarked video are performed.

The attacks are salt and pepper noise with various

densities (0.02,0.2), rotation(-3degrees to 3 degrees),

scaling(50% to 150%), cropping ,Gaussian

smoothing with various window sizes(3,5,7),

histogram equalisation, sharpening, jpeg

compression, video compression(mpeg, x264) at

various bit rates. Watermark is extracted from each

attacked video and both average PSNR on

watermarked videos and average number of bits in

error are calculated from each attacked video.

Eventually, each firefly value is calculated by using

objective function as given in Eq. (3).

𝑂𝑏𝑗 = 𝑃𝑆𝑅𝑁 + ∅ ∑ 𝐴𝑁𝐵𝐸(𝑊𝑀, 𝑊𝑀𝑖)𝑖=15𝑖=0 (3)

Where ANBE is average bits in error on given video,

which is the average of number of bit errors over the

video frames. WM is the original secret shared

watermark, WMi is the extracted watermark from ith

attacked video, when i=0 means no attack is

performed. Φ is the weight factor which will be

learned by optimisation and PSNR is defined as the

peak signal to noise ratio and is computed as given

in Eq. (5).

5. Experimental results

In this section, we demonstrate experimental

results of the proposed video watermarking method

by implementing it in MATLAB and measure

performance against a dataset comprises of 140 videos.

We also proved this with simulation results, how

chaotic firefly algorithm shows its influence on

video watermarking to compromise imperceptibility

and robustness along with watermark embedding

and extraction approach. We did simulations by

setting the watermark size to 15x20 and then secret

sharing image size becomes double than that of



watermark, that is 30x40, which represents a unique

logo assigned to the customer for tracking from

where the video is pirated or owner to prove

ownership and this is inserted into the video in avi

format of size 640x480 using the proposed method.

For generating the secret share image, we followed

the table given in [24]. Any format of a video can

be converted into avi format required for embedding

the watermark and the watermarked video is

converted back to original video by the FFMPEG

tool. Compromise between robustness and

imperceptibility is achieved by optimizing the

scaling factor in the interval of 10 and the optimized

parameter is computed using chaotic firefly

algorithm and fixed to 13. The performance of the

proposed method is tested on various kinds of video

genre. Perceptual visual quality is measured by

computing Peak Signal to Noise Ratio (PSNR) as

given in below equation and robustness of the

method is validated by calculating the number of

bits that are in error and then calculated error rate on

various kinds of video genre. We also evaluated the

performance and robustness of the method against

several attacks like median filtering, compression,

salt and pepper noise, rotation, scaling, frame drop,

frame swap and combination of these.

𝑀𝑆𝐸 = 1

𝑚𝑛 ∑ 𝑚−1

𝑖=1 ∑ (𝐼(𝑖, 𝑗) − 𝐾(𝑖, 𝑗))𝑛−1𝑗=1 (4)

𝑃𝑆𝑁𝑅 = 20 𝑙𝑜𝑔 (𝑀𝐴𝑋𝑡

𝑀𝑆𝐸 (5)

Where I, K are the input image and watermarked

image respectively. In this experiment, we

measured average PSNR means average of PSNR

values of videos frames of the watermarked video.

Similarly, we measured average number of bits in

error means average of number of bits in error frame

all the video frames of the watermarked and

attacked video. The initial chaotic firefly parameters

are set α is 1.0 and γ is 0.01. The Maximum number

of iteration is set at 15 and the firefly populations of

CFA are 15. The parameter ϕ in 10. To run chaotic

firefly algorithm we selected the scaling factors

from 10 to 25 and few attacks which were mention

earlier. For this we selected sports video as well as

news video.

Sample input logo that we considered in this

experiment is shown in following Fig. 5 and Fig. 6

shows the scrambled watermark logo shown in Fig.

5, this scrambling is done avoid burst errors. In our

case scrambling is done after secret share image

generation.

Figure.5 A watermark

Figure.6 Scrambled output of a watermark

(a) (b)

Figure.7 Share images: (a) generated from video frame1

and (b) generated from video frame 2

This input watermark logo is used to generate

shared image along with input video frame, it varies

from frame to frame and the same is inserted into

the corresponding frame as a watermark. Two

sample share images from two input video frames

are shown in Fig. 7. We have also shown the

experimental results of proposed method. First of all

quality of the video is tested by measuring PSNR

metric that ranges from 44 to 47 depending on video

genre for this method. This PSNR [26] is enough to

say that this method is acceptable for video

watermarking purpose to prove ownership.

Further the robustness can be increased by

increasing scaling factor at the cost of quality in

sense increase in scaling factor results in decrease

in PSNR while increase in retrieval accuracies and

the same is shown in Figs.8 and 9, respectively.

We also simulated various manipulations over

video(generally these can be termed as attacks)

which generally happen while transferring the video

over INTERNET, attacks such as rotation, scaling,

compression, frame rate conversion and frame drop

attack. Various compression methods like mp4

compression, mpeg compression etc. are also

performed to prove the robustness of the system.

Few of the attacks are simulated using MATLAB

and compression attacks are simulated using

FFMPEG [25]. The following Fig.10 shows the

extracted watermark when no attack occurs from

one frame of the avi video.



Figure.8 PSNR vs scaling factor

Figure.9 Retrieval accuracy vs scaling factor

Figure.10 Extracted watermark from one frame

Figs. 11 and 12 show when salt and pepper noise

is added to the video frames and extracted

watermark logo using proposed method when salt

and pepper noise with density is 0.02. In this

experiment, we achieved the retrieval accuracy of

99.333 percept.

The Table 1 shows the watermark retrieval

accuracy in terms of number of bits in error verses

various attacks and also compared with the one of

the state of art method [27] and [28]. The results

state that proposed method is able to show the

improvement in number of bits in error. Considered

one frame from the video to compare with state of

art.

We also evaluated the method by calculating

average PSNR to measure imperceptibility and

average number of bits in error to measure

robustness on various kinds of video genre when

rotation attack is -3 degrees. Same is illustrated in

Table 2.

Table 3 shows when the attack is mpeg

compression at bit rate of 1024kbps. PSNR will be

same in every case because it measured when

watermark is inserted. Only average bits in error are

changed.

Table 4 shows when the attack is x264

compression at bit rate of 2048kbps.

Figure.11 Salt and pepper noise is added with density of

0.02

Figure. 12 retried watermark with accuracy is

99.333

Table 1. Watermark method accuracy

Attack

Type

Number of Bits in Error

Method

[27]

Method

[28]

Proposed method

Rotation 2

degrees

16 9 6

Scaling

200%

6 3 1

Scaling

50%

4 2 1

Cropping

30%

12 5 8

MPEG 2

with

1024 Kbps

9 6 2

Frame

Drop (up to

400

consecutive

frames)

Not

able to

sync

0 0



Table 2. PSNR and average bit errors on various video

gerne when rotation is -3

Video Types PSNR(db) Avg. Bit Error

Sports Videos

(foot boll)

46.23 9.23

General Movie 46.13 3.45

Cartoon Video 45.23 2.56

News Video 44.9 5.21

Natural Video

(Geographic Channel)

46.89 6.03

Table 3. PSNR and average bit errors on various video

gerne when mpeg compression at 1024 kbps

Video Types PSNR(db) Avg. Bit

Error

Sports Videos(foot boll) 46.23 10.24




Natural Video


46.89 8.56

Table 4. PSNR and average bit errors on various Video

gerne when mpeg compression at 1024 kbps

Video Types PSNR(db) Avg. Bit

Error

Sports Videos (foot b0ll) 46.23 12.23




Natural Video


46.89 9.23

We also observed that 40% reduction in errors

because of secret sharing method.

6. Conclusion and discussion

In this paper, we proposed a solution for video

watermarking for claiming ownership especially to

counter or prevent illegal online video sharing. This

method uses discrete wavelet transform which in

turn used singular value decomposition to get

singular values because of its robustness even

though small change in coefficient does not change

its signal characteristics, scrambling method is

adopted to remove burst errors and secret sharing is

chosen in order to correct the errors by sharing each

watermark bit to four bits and retrieved original bit

based on majority voting during retrieval process.

Selected four subbands to generate secret shared

watermark to make the method robust against high

pass and low pass filtering methods. Compromise

between robustness and quality is achieved by using

the chaotic firefly optimisation method for this

embedding and extraction scheme. As whole

watermark is inserted into single frame, this method

is robust against frame drop attack. Experimental

section proved the robustness against various attacks

while maintaining the watermarked video quality.

We simulated various attacks on videos by using

FFMPEG and MATLAB. Also the experimental

results proved the superior in terms of performance

of the proposed method when compared with state

of art methods on mentioned attacks. In future work,

we would like to improve the retrieval accuracy, try

to improve the PSNR and also try to reduce

embedding and extraction cost. Adding to the above

mentioned point we would like to adopt various

optimisation methods and want to compare among

them.

References

[1] A. Averbuch, D. Lazar, and M. Israeli, “Image

compression using wavelet transform and

multiresolution decomposition”, International

Journal of IEEE Transactions on Image

Processing, Vol. 5, No. 1, pp. 4-15, Jan 1996.

[2] B. C. Mohan and S. S. Kumar, “A robust image

watermarking scheme using singular value

decomposition”, International Journal of

Multimedia, Vol. 3, No. 1, pp. 715, 2008.

[3] R.A. Sadek, “SVD Based Image Processing

Applications: State of the Art, Contributions

and Research Challenges”, International

Journal of Advanced Computer Science and

Application, Vol. 3, pp.26-34, 2012.

[4] H. Liu, N. Chen, J. Huang, X. Huang, and Y.

Q. Shi, “A robust DWT-based video

watermarking algorithm”, In: Proc. of IEEE

International Symposium on Phoenix-

Scottsdale, AZ, pp. 631-634, 2002.

[5] S. A. Rathore, S. A. M. Gilani, A. Mumtaz, T.

Jameel, and A. Sayyed, “Enhancing

Invisibility and Robustness of DWT based

Video Watermarking scheme for Copyright

Protection”, In: Proc. of International Conf. on

Information and Emerging Technologies,

Karachi, pp. 1-5, 2007.

[6] H. Tian and W. Ji, “A Digital Video

Watermarking Scheme Based on 1D-DWT”,

In: Proc. of International Conf. On Biomedical

Engineering and Computer Science, Wuhan, pp.

1-3, 2010.

[7] C. Wang, C. Zhang, and P. Hao, “A blind

video watermark detection method based on

3D-DWT transform”, In: Proc. of IEEE

International Conf. On Image Processing, pp.

3693-3696, 2010.



[8] C. W. H. Fung and W. Godoy, “A novel DWT-

SVD video watermarking scheme using side

view”, In: Proc. of 5th International Conf. On

Signal Processing and Communication Systems,

pp. 1-4, 2011.

[9] A. El Allali, J. Elabbadi, and E.I. Elahaj,

“Video object watermarking using 3D-Walsh

Hadamard transform and Arnold transform”,

In: Proc. of International Conf. On Multimedia

Computing and Systems, Tangier, Morocco, pp.

119-124, 2012.

[10] N. Dey, P. Das, and A. B. Roy, “DWT-DCT-

SVD based intravascular ultrasound video

watermarking”, In: Proc. of International Conf.

on Information and Communication

Technologies, Trivandrum, India, pp. 224-229,

2012.

[11] P. Prathik, R. Krishna, R. A. Nafde, and K.

Shreedarshan, “An Adaptive blind video

watermarking technique based on SD-BPSO

and DWT-SVD”, In: Proc. of International

Conf. On Computer Communication and

Informatics, Coimbatore, India, pp. 1-15, 2013.

[12] R. Maharjan, A. Alsadoon, P. W. C. Prasad, A.

M. S. Rahma, A. Elchouemi, and S. A.

Senanayake, “A Proposed Robust Video

Watermarking Algorithm: Enhanced Extraction

from Geometric Attacks”, IEEE International

Conf. on Multimedia and Image Processing,

Bandar Seri Begawan, Brunei, pp.45-50, 2016.

[13] Li and A. Sui, “A Digital Video Watermarking

Algorithm Based on DCT Domain”, IEEE

International Joint Conf. on Computational

Sciences and Optimization, Harbin, China, pp.

557-560, 2012.

[14] W. Trabelsi and M. H. Selmi, “Multi-signature

robust video watermarking”, IEEE

International Conf. On Advanced Technologies

for Signal and Image Processing (ATSIP),

Sousse, Tunisia, pp. 158-163, 2014.

[15] M. Sundararajan and G. Yamuna, “DWT based

scheme for video watermarking”, IEEE

International Conf. on Communication and

Signal Processing, Melmaruvathur, India, pp.

460-464, 2013.

[16] S. Kadu, C. Naveen, V. R. Satpute, and A. G.

Keskar, “Discrete wavelet transform based

video watermarking technique”, International

Conf. on Microelectronics, Computing and

Communications (MicroCom), Durgapur, India,

pp. 1-6, 2016.

[17] F. Akhlaghian and Z. Bahrami, “A new robust

video watermarking algorithm against cropping

and rotating attacks”, International Iranian

Society of Cryptology Conf. on Information

Security and Cryptology, Rasht, Iran, pp. 122-

127, 2015.

[18] S.L. Hsieh, J.J. Jian, I.J. Tsai, and B.Y. Huang,

“A color image watermarking scheme based on

secret sharing and wavelet transform”, In: Proc.

of IEEE International Conf. on Systems Man

and Cybernetics, Montreal, Que., Canada, pp.

213-214, 2007.

[19] M. Ali, “An introduction to wavelets and haar

transform”, http://www.cs.ucf.edu/mali/haar.

[20] K.H. Talukder and K. Harada, “Haar Wavelet

Based Approach for Image Compression and

Quality Assessment of Compressed Image”,

IAENG International Journal of Applied

Mathematics, Vol.36, pp.1-8, 2007.

[21] W. Kong, B. Yang, D. Wu, and X. Niu, “SVD

Based Blind Video Watermarking Algorithm”,

In: Proc. of First International Conf. on

Innovative Computing, Information and

Control, Beijing, China, pp.265-268, 2006.

[22] L. Dos Santos Coelho, D. L. de Andrade

Bernert, and V. C. Mariani, “A chaotic firefly

algorithm applied to reliability-redundancy

optimization”, IEEE Congress of Evolutionary

Computation (CEC), New Orleans, LA, pp.

517-521, 2011.

[23] R. Rhine and N.T. Bhuvan, “Image Scrambling

Methods for Image Hiding: A Survey”,

International Journal of Computer Science and

Information Technologies (IJCSIT), Vol. 5, pp.

751 – 755, 2014.

[24] S.L. Hsieh, J.J. Jian, I.J. Tsai, and B.Y. Huang,

“A color image watermarking scheme based

on secret sharing and wavelet transform”, In:

Proc. of IEEE International Conf. On Systems,

Man and Cybernetics, Montreal, Que, Canada,

pp. 2143-2148, 2007.

[25] https://ffmpeg.org/.

[26] Q.Huynh-Thu and M.Ghanbari, “Scope of

validity of PSNR in image/video quality

assessment”, Electronics Letters. Vol. 44, No.

13, pp.800-801, 2008.

[27] J. Li, P. Zhong, Y. Zhu, and C. Guo, “Robust

wavelet-based watermarking scheme for video

copyright protection”, In: Proc. of International

Congress on Image and Signal Processing

(CISP) , Dalian, China, pp.125-129, 2014.

[28] S.B. Latha, D.V. Reddy, and A. Damodaram,

“Digital Video Watermarking using DWT and

Singular Values”, In: Proc. of International

Conf. On Information and Communication

Technology for Competitive Strategies (ICTCS),

Udaipur, India, 2016.

Robust Video Watermarking Using Secret Sharing, SVD, … · Robust Video Watermarking Using Secret Sharing, SVD, DWT and Chaotic ... Despite many watermarking methods exist on combination

Documents